[FEA] Enable nvdashboard to show GPU memory utilization for multiple nodes

[randerzander]

With a Dask LocalCUDACluster it's easy to tell how well utilized my whole "cluster" of GPUs is.

For multi-node dask clusters where each node has GPUs, it would be great if nvdashboard could work with something like Dask's client.run API to gather the same utilization metrics from all nodes.

[jakirkham]

Related to issue ( #34 ) and issue ( #45 ).

[jakirkham]

I think this may already be supported by dask-labextension. Not sure if you have tried that already Randy. Please let us know if you are running into issues here and we can take a look 🙂

[jacobtomlinson]

Did you make any progress with the Dask lab extension @randerzander?

Nvdashboard is just for single node. Dask can show GPU metrics too which should solve your issue. One thing to note is that the lab extension only lists the GPU dashboards under certain conditions at the moment, which is frustrating.